help settle a methodology discussion.

[lre1234]

Hi all,

My boss and I are having a debate on a protocol that I need a little clarity on, and hopefully someone can give the best answer.

The problem: we are interested in using RNA-seq data and looking at strand specific expression data. But, all of the sequence data that we have is from paired-end RNA sequence.

My boss thinks that this isn't a problem and we can do one of two things. either a) only use the forward sequence reads and look at the data from that, or b) simply take the reverse reads, and use the reverse-compliment of them and treat them as forward reads.

My answer is that he is wrong in both cases. By doing paired-end sequencing, you inherently loose the strand information and cannot do any strand specific information. Simply reversing the two reads will not work, and you do not know which strand the forward (or reverse read came from). It is not always the case that the forward read is the plus strand and the reverse read is the minus strand, and these can be swapped. Essentially, we do not know where (strand and orientation) that the read came from.

Can someone help confirm that I am correct .

[dpryan]

You're both wrong actually. Whether strand information is conveyed is determined by the nature of the library prep, not whether you sequence single-end or paired-end. You can certainly do strand-specific paired-end sequencing, that's not even all that uncommon. So, the question is purely whether you have a strand-specific dataset. If so, then just determine which of the reads is conferring the strand information (normally read #2 does, since dUTP methods are the most popular) and proceed accordingly.

[dariober]

You're both wrong actually. Whether strand information is conveyed is determined by the nature of the library prep, not whether you sequence single-end or paired-end. You can certainly do strand-specific paired-end sequencing, that's not even all that uncommon. So, the question is purely whether you have a strand-specific dataset. If so, then just determine which of the reads is conferring the strand information (normally read #2 does, since dUTP methods are the most popular) and proceed accordingly.

As a side comment... Paired end sequencing is such a source of confusion. Sometimes I wish it was never invented!

[pmiguel]

As a side comment... Paired end sequencing is such a source of confusion. Sometimes I wish it was never invented!

Do you also wish you didn't have a nose every time you catch a cold?

--
Phillip

[pmiguel]

You're both wrong actually. Whether strand information is conveyed is determined by the nature of the library prep, not whether you sequence single-end or paired-end. You can certainly do strand-specific paired-end sequencing, that's not even all that uncommon. So, the question is purely whether you have a strand-specific dataset. If so, then just determine which of the reads is conferring the strand information (normally read #2 does, since dUTP methods are the most popular) and proceed accordingly.

Completely agree.

In addition I don't think informaticians should trust that data sets are of a particular strandedness. They should test data sets to determine that. Just align 1000 reads to the large rRNA subunit sequence for the species in question. If you have paired reads this will also answer the question "what insert size was sequenced".

--
Phillip

[dariober]

Do you also wish you didn't have a nose every time you catch a cold?

In case it wasn't obvious... My comment was ironic. I just wanted to point out that PE can make things confusing. (And if I really hated PE I would just use only the 1st-in-pair reads).

[pmiguel]

In case it wasn't obvious... My comment was ironic. I just wanted to point out that PE can make things confusing. (And if I really hated PE I would just use only the 1st-in-pair reads).

Hey, I'm not here to judge...

--
Phillip

[dariober]

Hey, I'm not here to judge...

--
Phillip

Sure, no worries!